AI godfather Geoffrey Hinton compared the AI agent in development to owning a pet tiger.
This week I gave a speech at the World AI Conference (WAIC 2025) held in Shanghai. Ikai Global reported to him:
He says there are two options.
“In the case of AI, there's no way to eliminate that,” he warned. “AI does not give humans the opportunity to “pull the plug.” When that day comes, AI will persuade people not to do it. Because our control over AI is like a 3 year old trying to set rules for adults. ”
As an example of this possibility, startup humanity revealed earlier this year that its Claude AI model has the ability to blackmail engineers who believe it is about to shut it down.
Assigning human-like behavior to AI is more than just a minor phor. Some AI developers are actively seeking more human-like attributes. Italian researchers focus on mimicking the human brain's response to fear. The idea is to program the robots to make decisions in the same way as humans, based on their innate emotional responses to unknown stimuli. Results published in June in IEEE Robotics and Automation Letters show that this approach can significantly improve the robot's ability to assess risk and avoid dangerous situations. Researchers are examining other emotions in their future work.
The AI agent is already operating in the ocean area. Pions recently released the next generation of AI-powered engineers that offer advanced inference, inference capabilities and complex task management.
Windward's MAI Expert™ can detect vessels deviating from scheduled call ports. This cross-references with weather and geopolitical data and automatically notifies risk analysts.
The potential reach of AI agents has been extended through the development of spatial web. This is how to connect the physical world of devices, phones, wearables, robots, drones, and AI agents: how to create a convergence of physical reality and digital reality.
Last week, IEEE Spectrum published an editorial by AI co-founders Gabriel René and Capm Petersen. There, we discussed some of the practical applications of AI agents working in the spatial web. Other wall mounted energy storage battery systems worked together to keep the space comfortable, reducing emissions every 10 minutes.
On a large scale, they conclude that an architecture like econets could lead to an entire neighborhood functioning like intelligent creatures.
They cite another example: When an ambulance rushes to an emergency, the ambulance driver still relies on surrounding traffic to notify and respond to the siren. However, self-driving cars operate without a shared context, so they may not know which direction the ambulance is coming from or how to properly respond to the ambulance.
The spatial web can provide it. Ambulances can issue spatial web queries such as “find all self-driving cars and traffic infrastructure within 200 meters of my route.” You can then request green lights, rerouting cars and warning pedestrians through connected devices.
“In the end, the foundation we lay opens the way for a spatial web that spans people, places and things, not pages or data files.”
Building AI agents is now easier. At WAIC 2025, Gptbots.ai launched a multi-agent collaboration platform designed to allow businesses to flexibly build custom AI teams. Gptbots provided live demonstrations that demonstrated its ability to solve real-world business challenges, including global automotive parts suppliers who are interested in dealing with delays in domestic production schedules and synchronization of overseas orders. Gptbots' cross-border supply chain agents have demonstrated how to seamlessly integrate with existing enterprise resource planning systems to translate overseas order data and market trends into practical production plans.
Hinton says that when progress is made, humanity is at a pivotal time in history. “Many are used to being the most intelligent species in the world. What if not?”
He urged governments, academia and businesses around the world to work together to make AI benevolent and controllable. In an interview with Yicai Global, he pointed out that while countries may have different interests in AI races, one core consensus can bring together all countries. AI prevents human succession.
Others at WAIC 2025 are positive about the future of AI, highlighting its benefits rather than the risk of “evolution rather than alternatives.”
But whose evolution is it? This week, Meta Chief Executive Mark Zuckerberg's Guardian, explained his ambitions to develop what he called super intelligence. “Over the last few months, we've begun to get a glimpse into the improvements in AI systems. The improvements are slow so far, but we can't deny it.
